CFD-based analysis and two-level aerodynamic optimization on graphics processing units

On the performance of a 2D unstructured computational rheology code on a GPU

“…The system of equations is assembled by sweeping through all edges of the computational mesh in order to collect contributions from each edge to both triangular control volume straddling that edge. However in parallel architectures it is necessary to adopt a special scheme to avoid the occurrence of race conditions [13,26], which occur when two or more threads (running in parallel) try to access simultaneously the same memory address, leading to information loss. One way to avoid this problem in parallel architectures is to adopt a colouring scheme when assembling the system of equations [13,26].…”

“…However in parallel architectures it is necessary to adopt a special scheme to avoid the occurrence of race conditions [13,26], which occur when two or more threads (running in parallel) try to access simultaneously the same memory address, leading to information loss. One way to avoid this problem in parallel architectures is to adopt a colouring scheme when assembling the system of equations [13,26]. In this work we coloured those edges that have independent diagonal contribution to the matrix which guarantees that each triangle receives only one contribution from each edge colour, i.e.…”

“…Corrigan et al [12] reported speed-ups of 7.4x using DP with an unstructured cell-centred mesh finite volume code (FVM), for solving the Euler equations. The most relevant speed-ups were obtained on the implementation of an unstructured mesh vertex-based FVM code developed by Kampolis et al [13], where maximum speed-ups of 28x (single precision), 25x (mixed precision) and 20x (double precision) for 2D and 3D NavierStokes solvers, were achieved. Subsequent work by the same group [14] improved their code and a maximum speed-up of 46x was achieved with the mixed precision technique, when solving unsteady laminar and turbulent flows.…”

Computation of the Isogeometric Analysis Stiffness Matrix on Gpu

“…A number of studies in engineering applications have been recently reported on a variety of GPU platforms using implicit computational algorithms [6]- [16]. Linear algebra applications have also been a topic of scientific interest for GPU implementations [17]- [20].…”

On the Common Conceptual and Computational Frameworks for Multiscale Modeling

“…Demanding fine-scale or multiscale simulations have been run successfully on grids 54 or HPC systems with arrays of CPUs and GPUs 55,56 , producing impressive results. Leveraging these advanced computing infrastructures, a special challenge to the multiscale modelling community is to develop software systems that effectively support the modellers in the development and execution of highquality simulation.…”